Improvements in substance use and social determinants of health associated with hepatitis C treatment initiation and sustained virologic response in opioid treatment programs

Highlights

• •We evaluated substance use and SDOH as predictors of HCV treatment and cure.
• •We correctly identified HCV treatment initiation in 75% and cure in 94%.
• •We grouped patients based upon substance use, SDOH, and medical comorbidities.
• •These approaches identify individual and regional-level differences.

Abstract

Background

Although opioid use disorder (OUD) and hepatitis C virus (HCV) frequently co-occur, few studies have evaluated substance use and social determinants of health (SDOH) that predict HCV treatment initiation and sustained virologic response (SVR).

Methods

We used administrative and clinical trial outcome data from HCV-infected methadone-treated participants ($n=522$) to evaluate changes in substance use and SDOH associated with HCV treatment initiation and SVR. We performed random forest analysis for investigating factors that predict HCV treatment initiation and SVR. Additionally, we applied the factor analysis for mixed data method for dimensionality reduction and used a clustering method to divide participants into subgroups to assess individual and regional-level differences.

Results

HCV treatment approach (i.e., through telemedicine or referral), methadone duration, age, and degree of substance use correctly identified HCV treatment initiation in 75% and SVR in 94% of participants, respectively. Grouping participants based upon SDOH (first principal dimension), substance use (second dimension), and medical comorbidities, mental health disorders and demographics (third dimension) identified individual and regional level differences.

Conclusions

We identified changes in substance use and SDOH associated with HCV treatment initiation and SVR. Individual and regional-level subgroup differences could guide intervention implementation. HCV treatment initiation and SVR lead to improvements, not only in liver and infectious diseases outcomes, but in social and substance use factors.

1. Introduction

People with opioid use disorder (OUD) have the highest incidence and prevalence of hepatitis C virus (HCV). In the United States, approximately 2.2 million individuals have OUD (Jones et al., 2023) and 2.4 million individuals are HCV-infected (Ryerson et al., 2020). The annual HCV incidence per 100,000 individuals in the US increased four-fold, from 0.3 in 2009 to 1.2 in 2018 (Ryerson et al., 2020). Globally and in the Americas, pooled HCV incidence was 12.1 per 100 person-years between 1992 and 2021 among people who inject drugs (Artenie et al., 2023). While underreported, the estimated HCV prevalence among people who inject drugs ranges from 70 to 90% (Scarpetta et al., 2025). Unfortunately, while people with OUD are the key population to be targeted for HCV elimination, they are often disenfranchised from healthcare systems.

In the treatment of OUD, integrating behavioral and medical services can improve patient outcomes, social functioning, and healthcare costs; while separating these services leads to inferior care (Bornstein, 2020). Healthcare access for people with OUD is often reduced due to stigma, provider shortages, low patient priority, and inconvenience (Biancarelli et al., 2019, Harris et al., 2021, Talal et al., 2023, Zeremski et al., 2013). Opioid treatment programs (OTPs) offer methadone and behavioral therapy to manage OUD (Simon et al., 2022). OTPs also provide extensive support for HCV management through the provision of structure, purpose, and safe environments (Talal et al., 2021, Talal et al., 2023). Integration of HCV management into OTPs is a potential alternative to expand access to testing and treatment (Ghany et al., 2020), that may foster a therapeutic environment, patient engagement, and retention in healthcare (Islam, 2013). Integrated OUD-HCV care is evidence-based as it improves treatment outcomes and adherence (Artenie et al., 2020, Bartlett et al., 2022, Rosenthal et al., 2020, Springer and Del Rio, 2020), while circumventing stigma faced in conventional healthcare settings. We have recently completed a randomized controlled trial (RCT) of facilitated telemedicine for HCV management integrated into OTPs compared with offsite referral (Talal et al., 2024). While HCV treatment outcomes, including treatment initiation and sustained virological response rates (SVR), were two-fold higher in individuals treated through telemedicine, we also noted improvements in substance use parameters. Among all participants who achieved an SVR or HCV cure, we noted significant decreases in substance use as assessed by the drug abuse screening test (DAST-10) (Skinner, 1982), and minimal reinfections during 2 years of follow-up. These findings suggest a potential association between an HCV cure and decreased substance use.

As OUD affects all aspects of patients’ health and lives, many OTPs offer social services and psychological support to their clients to promote substance use recovery. Limited evidence exists on the role of social determinants of health (SDOH) and aspects of substance use treatment as predictors of HCV treatment initiation and cure (Hernandez-Vallant and Hurlocker, 2024, Pham et al., 2024). Using clinical and SDOH data collected from OTPs, we sought to understand the impact of improvements in OUD treatment outcomes, SDOH, and medical co-morbidities as predictors of HCV treatment initiation and SVR. Additionally, we grouped participants with the aim to identify individual and regional-level differences. These groupings were interpreted for clinical significance. An enhanced understanding of regional differences in SDOH could improve not only HCV treatment outcomes, but those related to OUD as well (Pham et al., 2024).

2. Methods

2.1. Parent study overview

The parent study was a prospective, cluster RCT using a stepped wedge-design (Talal et al., 2024). The objective was to compare SVR through facilitated telemedicine integrated into 12 OTPs in New York State to offsite referral. Facilitated telemedicine is videoconferencing between a patient and a remote hepatitis specialist, supported by a case manager, onsite phlebotomy, and HCV medication dispensing. We enrolled 602 participants from March 2017-February 2020. The inclusion criteria were >18 years old, 6-months of active OTP enrollment, HCV-treatment-naïve, and active coverage by health insurance. The exclusion criteria was HCV treatment ineligibility for any reason, most importantly undetectable HCV RNA. There were no inclusion/exclusion criteria based upon HCV genotype or fibrosis stage. We achieved SVR in 90% of participants through facilitated telemedicine as compared to 39% through offsite referral. In addition, we noted significant decreases in substance use among all cured participants and minimal reinfections during 2 years of follow-up.

2.2. Study design, data acquisition, and patient consent statement

The vast majority of data used in the analyses presented in this manuscript were retrospective data on substance use and SDOH collected by the OTPs that participated in the parent study. These data were combined with data from the RCT that provided the structure for patient enrollment. The parent study was approved by the Institutional Review Board at the University at Buffalo, and participants provided written informed consent. OTPs in New York State collect essential information both at admission (i.e., admission questionnaire) and annually for as long as an individual remains actively enrolled in the OTP (i.e., annual update surveys). People with OUD may have multiple OTP admissions. For participants with multiple OTP admissions, we utilized the most recent admission questionnaire obtained before HCV treatment initiation and annual update surveys obtained up to 4 years prior to RCT enrollment and every subsequent year during RCT participation (eFig. 1). Table 1 presents a list of variables used in the analysis, and study participants are described in eAppendix 1. The various steps in the data collection process are outlined in eAppendix 2. A glossary comprising of definitions of relevant terms to this work are provided in eTable 1. Additionally, eFig. 1 presents a graphical overview of the steps involved in data collection and analysis. eAppendix 3 outlines the steps in the data alignment process and construction of the cumulative percentage plots. The data used in this study were derived from admission and annual update forms. Admission data from the participants were collected between March 5, 1984 and November 29, 2019, while annual update data considered in the study were collected between April 16, 2013 and December 14, 2021.

Table 1.

Analytical Variables. Longitudinal variables are summarized via the improvement index.

Variable type	Variable Name
Cross Sectional	HCV treatment approach
	Sustained virological response
	Hepatitis C treatment initiation
	Months in methadone program
	Drug abuse screening test-10
	Area (urban/rural)
	Race
	Ethnicity
	Sex
	Age
	Comorbid conditions
	Ever treated for mental illness
	Inability to obtain phone
	Inability to obtain healthcare
	Inability to obtain childcare
	Inability to obtain utilities
	Inability to obtain food
	Ever been in jail in the past year
	Worried about losing housing
	Inability to obtain stable housing
	Marital status
	Inability to obtain clothing
	Age of primary substance use
	Route of primary substance
	Age of second substance use
	Child of someone who misuses alcohol or other substances
	Primary income at admission
Longitudinal-Improvement indices	Primary substance
	Primary frequency
	Secondary substance
	Secondary frequency
	Tertiary substance
	Tertiary frequency
	Criminal justice
	Employment
	Education
	Type of residence

The variables exhibit binary, categorical, and interval/ratio measurement scales. Longitudinal variables are summarized nonparametrically via improvement indices as described in Section 2.4. Abbreviations: hepatitis C virus, HCV:

2.3. Descriptive and preliminary analysis

For descriptive and exploratory analyses, cross-sectional nominal variables are summarized as absolute counts and relative frequencies. Continuous variables are summarized either as mean and standard deviation (SD) or median and interquartile ranges (IQR). All descriptive and preliminary analyses were performed using the R programming language (Version 4.2.3) (R Core Team, 2022).

2.4. Feature extraction from longitudinal variables

To quantify the improvement in a participant's various social and substance use-related factors over time, we develop a measure called the “improvement index”. The index summarizes, in a nonparametric way, the longitudinal variables associated with each participant. In the context of substance use, an improvement might be a transition from active substance use to complete abstinence. Similarly, a reduction in frequency of substance use, such as moving from daily use to weekly use, is also considered an improvement. An improvement in criminal status is identified when a participant transitions from any involvement with the criminal justice system (e.g., probation, parole, or serious legal issues) to not being involved with the criminal justice system. Educational improvement is noted when a participant advances to a higher academic level (e.g., from elementary school to high school). Finally, an improvement in employment status is recognized when a previously unemployed participant becomes employed or is classified as not in the labor force due to other legitimate reasons (i.e., retirement, disability, or participation in OUD treatment as mandated by New York State). We considered a change in any factor to have occurred only when a participant's response varied between successive forms.

The improvement index is defined as i = p/(n−1) where p denotes the number of times where the response improved between two successive forms and n is the total number of forms completed by a participant. This calculation normalizes the data to account for variations in the number of forms collected per participant. Subtracting one acknowledges the fact that a participant could have potentially changed their response (n-1) times if they completed n forms. The improvement index ranges from 0 to 1. A score of 0 signifies a complete lack of change. Conversely, a score of 1 suggests a high degree of consistent improvement over the course of the period of evaluation.

We have also formulated procedures to handle borderline cases, that is, participants with only one collected form and missing values. For a participant with one collected form, the improvement index is set to zero. If there is a missing value at time point i, we measure the change between time points i-1 and i + 1, assuming consistency between two consecutive time points. If all the recorded data for a variable are missing, the index is assigned as having a missing value.

2.5. Missing value imputation

Missing value imputation was performed using missForest. Details on missing values imputation are provided in eAppendix 4. Variables used for missing value imputation are illustrated in eTable 2.

2.6. Random forests (RF) application

We used RF to evaluate the predictive ability of the selected variables on HCV treatment initiation and SVR. HCV treatment initiation and SVR are represented as binary values, where 1 indicates the occurrence of either HCV treatment initiation or SVR, and 0 signifies their absence. When HCV treatment initiation is the outcome variable, we used data collected before the treatment start date, resulting in a total of 1584 forms. When SVR is the outcome, data up to the date of SVR achievement is used, providing 1749 forms to consider. The complete list of variables included in the random forest predictive model when using either HCV treatment initiation or SVR as outcomes along with the associated out-of-bag error is depicted in Table 2. We used Python, version 3.9.7 (Python Software Foundation, 2021) for performing the random forests classifications (5000 trees were employed) using the scikit-learn, Version 1.1.2 (Breiman, 2001, Pedregosa et al., 2011).

Table 2.

Random Forest Modeling. The out-of-bag errors and the variables utilized for predicting “treatment initiation” and “sustained virological response (SVR)” status through random forest models.

Predicted Outcome	Number of Participants	Variables Used	Out-of-bag Error
Treatment Initiation	522	Arm, Demographic data, improvement indices related to primary substance frequency, primary substance use, employment, secondary substance use, education, criminal justice status, and type of residence, DAST-10 scores upon study enrollment, participants' enrollment duration at the OTP, presence of comorbidities, marital status, as well as an array of social variables such as housing, nutrition, utilities, healthcare, essential items, phone accessibility, childcare, and incarceration.	0.249
SVR	345	Arm, Demographic data, DAST-10 scores upon study enrollment, participants' enrollment duration at the OTP, presence of comorbidities, marital status, as well as an array of social variables such as housing, nutrition, utilities, healthcare, essential items, phone accessibility, childcare, and incarceration.	0.055

Abbreviations: DAST-10, drug abuse screening test-10; OTP, opioid treatment program; SVR, sustained virological response.

2.7. Factor analysis for mixed-type data (FAMD) followed by k-means clustering

In a preliminary study, Williams et al. investigated the influence of SDOH and substance use factors on HCV treatment initiation using multiple correspondence analysis (Williams et al., 2019). Here, we used FAMD to build upon the prior work with a much larger sample, incorporating longitudinal data to understand how strongly and in what way substance use and SDOH variables were synergistic. Since the data were of “mixed-type” (i.e., interval/ratio and categorical/nominal), we used FAMD to perform dimensionality reduction of the variable space, and return a set of numerical variables in a three-dimensional space before applying k-means clustering on the derived data.

FAMD and k-means analyses have been executed using R version 4.2.3. FAMD has been performed using the FAMD function of the FactoMineR package (Lê et al., 2008); the k-means (R function “kmeans” from the stat package) algorithm was used for clustering participants into four clusters (eFig. 2). Please note that when discussing the methods or results of FAMD analysis, we use the words “cluster” and “subgroup” interchangeably.

We retained the first three principal dimensions that explained 26.1% of the variance due to their explainability, parsimony, and ease of interpretation. The first principal dimension, explaining 10.6% of the variation, reflects SDOH (i.e., food, clothing, utilities, healthcare, phone, and childcare). The second principal dimension, explaining 8.5%, reflects substance use and frequency. The third dimension, explaining 7.0%, encompasses medical comorbidities, demographics, and mental illness. To decide the optimal number of clusters, we used the elbow plot that depicts the number of clusters (x-axis) versus the value of the total within-cluster sum of squares. Next, we used the selected principal dimensions and performed k-means clustering to divide participants into four subgroups (eFig. 2).

3. Results

3.1. Number of participants and forms

From 522 participants, we evaluated a median of 4 (IQR [3, 5], range 1 to 8) forms per participant, of whom 48 (9%) had only an admission questionnaire without any annual updates.

3.2. Demographics

Most participants were male (62.3%) with a mean age of 48.6 ± 13.0 years, with substantial representation of African American (22.8%) and Hispanic (31.8%) individuals (Table 3). The most common comorbidities were mental health disorders, occurring in 22.4% of participants, while 39.5% of participants had no comorbidities. Most (59%) participants had never been married, the majority (76.8%) did not reside with children and had a median of 1 child per participant.

Table 3.

Demographic, Substance Use, Social and Improvement Index Variables from Participants (N = 522).

Variable Types	No. (%)
Age
Mean (SD)	48.6 (13.0)
Median [Q1, Q3]	49.0 [38.0, 59.0]
Min – Max	22.0–––75.0
Sex
Female	197 (37.7%)
Male	325 (62.3%)
Ethnicity
Hispanic or Latina	166 (31.8%)
Race*
Black or African American	119 (22.8%)
Other	160 (30.7%)
White	243 (46.6%)
Geographic location
Urban	452 (86.6%)
Ever treated for mental illness
No	279 (53.4%)
Yes	243 (46.6%)
Marital Status
Divorced	58 (11.1%)
Living as married	33 (6.3%)
Married	53 (10.2%)
Never married	311 (59.6%)
Separated	39 (7.5%)
Widowed	28 (5.4%)
Number of children living with client
0	401 (76.8%)
1–6	92 ((17.6%)
Missing, n. (%)	29 (5.6%)
Number of children
Mean (SD)	1.7 (1.8)
Median [Q1, Q3]	1.0 [0, 3.0]
Min − Max	0 – 9
Missing, n. (%)	9 (1.7%)
Comorbidity
Both mental and physical	101 (19.3%)
Mental	117 (22.4%)
None	206 (39.5%)
Other	98 (18.8%)
Hepatitis C treatment approach
Referral	257 (49.2%)
Telemedicine	265 (50.8%)
Hepatitis C virus treatment initiation
No	177 (33.9%)
Yes	345 (66.1%)
Sustained virologic response
No	196 (37.5%)
Yes	326 (62.5%)
Ever been in jail in the past year
No	473 (90.6%)
Yes	49 (9.4%)
Primary income at admission
Alimony/child support	3 (0.6%)
Department of Veterans Affairs	3 (0.6%)
Fam and/or spouse contribution	57 (10.9%)
None	84 (16.1%)
Other	51 (9.8%)
Safety net assistance (SNA)	102 (19.5%)
Supplemental Security Income/Social Security Disability Benefits	134 (25.7%)
Temporary Assistance for Needy Families (TANF)	21 (4.0%)
Wages/Salary	62 (11.9%)
Missing	5 (1.0%)
Substance Use Variables
Medication assisted opioid therapy
No	4 (0.8%)
Yes	345 (66.1%)
Missing	173 (33.1%)
Months in methadone program
Mean (SD)	54.4 (65.9)
Median [Q1, Q3]	27.0 [12.0,72.0]
Min − Max	4.0–––500
Drug abuse screening test −10**
Mean (SD)	4.66 (3.11)
Median [Q1, Q3]	5.00 [2.00,7.00]
Min – Max	0–10.0
Currently attending substance use self-help group meeting
No	390 (74.7%)
Yes	57 (10.9%)
Missing	75 (14.4%)
Gambling screen positive
No	284 (54.4%)
Not screened	154 (29.5%)
Yes	9 (1.7%)
Missing	75 (14.4%)
Prior substance use treatment episodes
Mean (SD)	2.7 (1.9)
Median [Q1, Q3]	2.0 [1.0, 5.0]
Min – Max	0–9.0
Missing	10 (1.9%)
Primary age (Age at which primary substance was first used)
Mean (SD)	22.3 (9.0)
Min – Max	1.0–99.0
Primary route
sInhalation	162 (31.0%)
Injection	331 (63.4%)
Oral	24 (4.6%)
Other	3 (0.6%)
Smoking	1 (0.2%)
Missing	1 (0.2%)
Secondary age (Age at which secondary substance was first used)
Mean (SD)	21.85 (9.743)
Min – Max	8.0–64.0
Second substance not present	179 (34.3%)
Child of someone who misuses alcohol or other substances
Child of someone who misuses alcohol	93 (17.8%)
Child of someone who misuses both alcohol and other substances	66 (12.6%)
Child of someone who misuses other substances	43 (8.2%)
No	315 (60.3%)
Missing	5 (1.0%)
Social Variables
Inability to obtain stable housing
No	360 (69.0%)
Yes	62 (31.0%)
Worried about losing housing
No	392 (75.1%)
Yes	130 (24.9%)
Inability to obtain food
No	439 (84.1%)
Yes	83 (15.9%)
Inability to obtain clothing
No	444 (85.1%)
Yes	78 (14.9%)
Inability to obtain utilities
No	457 (87.5%)
Yes	65 (12.5%)
Inability to obtain childcare
No	507 (97.1%)
Yes	15 (2.9%)
Inability to obtain healthcare
No	463 (88.7%)
Yes	59 (11.3%)
Inability to obtain phone
No	439 (84.1%)
Yes	83 (15.9%)
Inability to obtain other essential items
No	497 (95.2%)
Yes	25 (4.8%)
Improvement Indices
Primary substance improvement index
Mean (SD)	0.2 (0.3)
Median [Q1, Q3]	0 [0.0, 0.3]
Primary frequency improvement index
Mean (SD)	0.3 (0.3)
Median [Q1, Q3]	0.3 [0.0, 0.5]
Secondary substance improvement index
Mean (SD)	0.2 (0.3)
Median [Q1, Q3]	0 [0.0, 0.3]
Secondary frequency improvement index
Mean (SD)	0.3 (0.3)
Median [Q1, Q3]	0.2 [0.0, 0.4]
Tertiary substance improvement index
Mean (SD)	0.1 (0.2)
Median [Q1, Q3]	0 [0.0, 0.2]
Tertiary frequency improvement index
Mean (SD)	0.2 (0.3)
Median [Q1, Q3]	0 [0.0, 0.3]
Criminal justice status improvement index
Mean (SD)	0.04 (0.1)
Median [Q1, Q3]	0 [0.0, 0.0]
Type of residence improvement index
Mean (SD)	0.08 (0.21)
Median [Q1, Q3]	0 [0.0, 0.0]
Employment improvement index
Mean (SD)	0.15 (0.23)
Median [Q1, Q3]	0 [0.0, 0.3]
Education improvement index
Mean (SD)	0.05 (0.14)
Median [Q1, Q3]	0 [0.0, 0.0]

Abbreviations: SD, standard deviation; IQR, interquartile range, Min, minimum; Max, maximum.

*Race assigned according to five standard categories from the National Institutes of Health (https://grants.nih.gov/grants/guide/notice-files/not-od-15-089.html).

**Drug Abuse Screening Test-10 (DAST-10) score ranges from 0 to 10 where a score from 3 to 5 represents a moderate level of issues related to substance use (Skinner, 1982).

3.3. Longitudinal changes

Longitudinal changes in substance use and SDOH before and after HCV treatment initiation are described in eAppendix 5. eFig. 3 presents the cumulative percentage plots of longitudinal categorical variables.

3.4. Variables associated with HCV treatment initiation and SVR

Among the 522 study participants, the most important variables associated with treatment initiation were, in descending order of importance: HCV treatment approach, age, methadone duration, and DAST-10 score (Fig. 1A). Treatment approach (either through telemedicine or offsite referral) accounted for 21% of the importance, while collectively age, median methadone duration, and DAST-10 score accounted for 24% of the importance. The next most influential variables were improvements in substance use and frequency as well as employment. These variables correctly identified HCV treatment initiation in about 75% of participants.

Fig. 1.

Random forests feature importance based on the Gini importance or Mean Decrease in Impurity (MDI) (Nembrini et al., 2018) Fig. 1A: The feature importance plot using MDI when hepatitis C virus (HCV) treatment initiation is predicted for 522 participants, using the HCV treatment approach, demographic information, comorbidities, marital status, improvement indices and SDOH associated with housing, nutrition, utilities, healthcare, essential items, phone access, childcare, and incarceration. Treatment approach, age, number of months on methadone, drug abuse screening test (DAST) 10 score at study enrollment, and improvement indices related to addiction and employment were the most important variables associated with HCV treatment initiation. Fig. 1B: The feature importance plot when sustained virological response (SVR) is predicted for 345 participants who initiated treatment using HCV treatment approach, improvements in social and substance use measures, such as primary substance frequency, primary substance use, employment, secondary substance use, education, criminal justice status, type of residence, demographic data, DAST 10 scores upon study enrollment, participants' enrollment duration on methadone, presence of comorbidities, marital status, as well as SDOH related to housing, nutrition, utilities, healthcare, essential items, phone accessibility, childcare, and incarceration. The number of months on methadone, age, DAST 10 score at study enrollment and HCV treatment approach were found to be the most important variables. The improvement indices related to substance use and employment were the next most important variables for predicting SVR.

Among the 345 participants who initiated HCV treatment, the most important variables associated with SVR were the same as those for HCV treatment initiation although of different levels of importance. The most important variable was methadone duration followed by age, DAST-10 score, and HCV treatment approach (Fig. 1B). Improvements in employment, criminality, and substance use and frequency were the next most influential variables. These variables correctly identified achieving an SVR in 94% of participants.

3.5. Results of the FAMD analysis and k-means clustering

Dividing participants into subgroups permitted evaluation of interrelationships between variables at individual and regional levels (eAppendix 6). The overall clustering results are illustrated in eTable 3a and summarized in Fig. 2. HCV treatment approach (i.e., telemedicine or referral), treatment initiation, and outcome clustering results are shown in eTable 3b. We found that subgroup 1 had the greatest improvements in substance use, residence type, and employment, along with moderate DAST-10 scores, and the shortest median methadone duration. Subgroup 2 had the highest difficulty in obtaining food, clothing, utilities, childcare, healthcare, phone, housing, and essential items. It also had the highest DAST-10 scores and the second shortest median methadone duration. Subgroup 3 had the highest prevalence of medical and mental health comorbidities and the highest prevalence of prior treatment for mental illness. It also had the greatest improvement in criminality, and its members predominantly reside in urban locales. Subgroup 4 had the greatest improvement in education, and the lowest positive changes for substance use. Participants in this cluster had the longest methadone duration.

Fig. 2.

Heatmap of Clusters by Grouped Variables: The plot illustrates the characteristic patterns of the four identified clusters across various grouped variables. On the X-axis, clusters are displayed alongside their respective sizes. The Y-axis categorizes variables by their thematic relevance, including clinical trial metrics, demographics, disease prevalence, SDOH, and improvement indices. Each category is demarcated by black horizontal lines. Within this matrix, the minimum and maximum values across clusters are represented by yellow and purple, respectively, with intermediate values shaded in light grey. When adjacent rows within a cluster share the same color, they are merged into a single-color block for visual simplicity. This plot emphasizes only the extreme (i.e., maximum and minimum) values, omitting intermediary data. For complete statistics for the clusters please refer to eTable 3a. Image symbols: Diamond denotes the variables for which clustering is based upon median, % denotes category percentages, and an Asterix denotes mean. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

To investigate the relationship between the identified clusters and the participants’ counties and regions of residence across New York State (Fig. 3), we ranked counties based upon percent in poverty (United States Census Bureau), and evaluated the number and percentage of individuals in each cluster stratified by New York State county (eTable 4). Additionally, we grouped the counties into five regions and evaluated the number and percentage of individuals per cluster per region (eTable 5). We note that approximately 59% of Cluster 4 participants originate from the seven counties where the percentage of individuals in poverty exceeds 15%. When evaluating the data based upon region, we note that the Downstate region has the highest poverty level and the greatest number of individuals without improvement in SDOH. The Hudson Valley has participants divided between clusters 3 and 4. The Finger Lakes has the lowest percentage of individuals in poverty with the highest percentage of participants in cluster 3.

Fig. 3.

New York State County map illustrating regions, study site locations, and participants per county. The figure illustrates the distribution of study participants per county according to the color scheme illustrated (right). Counties of participant origin are designated according to different regions as indicated by the boxes delineating each region. The location of each of the 12 opioid treatment programs that participated in the randomized controlled trial is indicated by a purple star. Map lines delineate counties of participant origin and do not necessarily depict accepted national boundaries. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

4. Discussion

We investigated approaches to maximize the effectiveness of integrated OUD and HCV treatment by identifying factors associated with HCV treatment initiation and SVR as well as performing individual- and regional-level subgroup classifications. Since people with OUD may require social support during HCV treatment, we assessed variables associated with engagement (i.e., HCV treatment initiation) and retention (i.e., SVR). We also clustered participants based upon demographics, comorbidities, SDOH, and improvement indices to evaluate differences in these variables at individual and regional levels. The most important variables associated with HCV treatment initiation were, in descending order of importance, HCV treatment approach, age, methadone duration, and DAST-10 score. Treatment approach accounted for 21% of the importance, while collectively age, methadone duration, and DAST-10 score accounted for 24%. The most important variables associated with SVR were the same as for HCV treatment initiation, although of different levels of importance; these factors are in descending order of importance: methadone duration, age, DAST-10 score, and HCV treatment approach. While HCV treatment should not be withheld from individuals with ongoing substance use (Artenie et al., 2020, Bartlett et al., 2022, Ghany et al., 2020, Rosenthal et al., 2020, Springer and Del Rio, 2020), the findings suggest that initiating HCV treatment once individuals have begun methadone may improve therapeutic outcomes. In a qualitative study, HCV treatment completers described their substance use as manageable, while non-completers described substance use-associated symptoms, such as decreased energy and motivation (Karasz et al., 2022). In agreement with our findings related to HCV treatment initiation and SVR, another study found that advanced age promoted HCV treatment adherence (Frankova et al., 2021). These results highlight potential benefits of integrating OUD and HCV treatment.

Clustering participants into subgroups permitted evaluation of interrelationships between variables at individual and regional levels. We used three principal dimensions: the first is exclusively comprised of SDOH, the second consists of substance use variables, and the third is a combination of medical comorbidities, mental health, and demographic variables. Subgroup 1 had the greatest improvements in substance use, residence type, and employment, and the shortest methadone duration, implying that methadone initiation improved subgroup 1 substance use and SDOH (housing and employment). Subgroup 2 had the highest DAST-10 scores and the greatest difficulty in obtaining food, clothing, utilities, childcare, healthcare, phone, housing, and essential items, implying that uncontrolled substance use is linked to deficiencies of essential needs. Subgroup 3 had the highest prevalence of medical and mental health comorbidities. It also had the highest proportion of HCV treatment initiation and SVR. Since subgroup 3 cluster members had other medical and mental health disorders, they were likely familiar and engaged with healthcare systems. The prior knowledge may have been permissive for successful navigation of the healthcare system for obtaining HCV treatment.

Subgroup 4 had the longest methadone duration and the lowest indices of positive changes for substance use. Most participants in this subgroup originated from the Downstate region that had high poverty levels. These findings imply that while regional poverty did not restrict access to OUD treatment, OUD outcomes could have been improved with support for social services. See eTable 4 in Supplementary material. Winkelman et al. found that OUD was associated with lower education levels and high rates of unemployment and criminality (Winkelman et al., 2018). In another national study, homelessness was significantly and negatively associated with OTP admission and retention (Gaeta Gazzola et al., 2023). Other risk factors for OTP discontinuation include younger age (i.e., 18–29 years), methamphetamine use, and referral from criminal justice centers, schools, employers, or community sources (Krawczyk et al., 2021). Our investigation begins to elucidate the relationship between substance use and SDOH on HCV and OUD outcomes. Given the importance of the number of months enrolled in the OTP and other SDOH as predictors of HCV treatment initiation and SVR, further evidence is required to understand the effects of SDOH on both HCV and OUD outcomes.

Consideration of nonclinical factors when evaluating HCV and OUD treatment outcomes is critical as they may account for up to 70% of health issues (Daniel et al., 2018). Building the nonclinical factors evidence base starts with collection of SDOH data. SDOH data collection in clinical settings may promote social service integration, which could be supported by clustering participants not only at the individual but also at regional levels (Kepper et al., 2023, Outland et al., 2022, Phuong et al., 2021, Wang et al., 2021). To investigate regional differences in SDOH and substance use, we evaluated the percentage of participants in each subgroup stratified by New York State County and region of residence. The regional differences analysis enabled us to evaluate the effect of county level poverty on changes in SDOH. Our results demonstrate, for example, that the Downstate region has the highest poverty level accompanied by the greatest number of participants with the least improvement in substance use. Clustering participants into subgroups and evaluating subgroups across different regions may facilitate more efficient resource allocation to promote support services for HCV and OUD treatment (Phuong et al., 2021, Weir et al., 2020). For example, social service integration is supported by recent work utilizing geospatial mapping that revealed mismatches between social service need and provision (Shadowen et al., 2022). Efficient resource allocation supports geographical proximity, which has been positively associated with methadone dispensing (Amiri et al., 2020). Social needs are heavily influenced by area-level vulnerability with high variability between domains dependent upon community-level characteristics (Brignone et al., 2024). Identifying mismatches between social service needs and provision may pinpoint regions where incentives, such as tax-breaks to social service organizations, could promote service delivery. Thus, the identification of individual and regional-level SDOH is important to improve overall health status.

4.1. Strengths and limitations

Strengths of this work include the collection of administrative data from reliably accurate and valid sources and its utilization for research purposes. Our approaches identify areas where social support is needed at both individual and regional levels. Furthermore, strengths of the work include innovative approaches for the analysis of complex data. Limitations of the investigation include the relatively small number of participants and limited data availability per participant. Our participants were enrolled in OTPs from New York State, the state provides more support for methadone dispensing than other states (Simon et al., 2022). Furthermore, individuals in methadone-based opioid use disorder treatment may not be representative of the larger substance use treatment population. Additionally, the administrative data may have omitted many empathy and trust-related variables, which might affect health status or an individual’s willingness to participate in programs to improve SDOH.

4.2. Conclusions

When evaluating predictors of HCV treatment initiation and SVR, we found that methadone duration and the HCV treatment approach (i.e., telemedicine or referral) were the most important. Additionally, our innovative analytical methods enabled comparisons between variables at individual and regional levels, which could guide implementation of patient-centered healthcare interventions. Few studies have evaluated SDOH and substance use variables as predictors of HCV treatment outcomes, especially in the context of the OTP. As access to HCV care is important, especially for resource-limited populations (Valdiserri et al., 2023), satisfactorily addressing substance use and SDOH that affect HCV treatment initiation and SVR in people with OUD is crucial to achieve HCV elimination goals.

Trial Registration

Clintrials.gov registration number NCT02933970.

Funding information

This work was supported by Patient-Centered Outcomes Research Institute (PCORI) Awards (IHS-1507-31640 [AHT] and ME-2024C1-37584 [MM]) and partially supported by the Troup Fund of the Kaleida Health Foundation (to AHT). The statements in this work are solely the responsibility of the authors and do not necessarily represent the views of PCORI, its Board of Governors or Methodology Committee. Study funders were not involved in data collection, analysis, or manuscript preparation.

CRediT authorship contribution statement

Andrew H. Talal: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Resources, Project administration, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Raktim Mukhopadhyay: Writing – review & editing, Writing – original draft, Visualization, Validation, Software, Methodology, Investigation, Formal analysis, Data curation. Valentina Veronesi: Writing – review & editing, Writing – original draft, Visualization, Methodology, Investigation, Formal analysis. Arpan Dharia: Writing – review & editing, Writing – original draft, Visualization, Investigation, Formal analysis. Giovanni Saraceno: Writing – review & editing, Writing – original draft, Visualization, Methodology, Investigation, Formal analysis. Daniel S. Fierer: Writing – review & editing, Writing – original draft, Visualization, Investigation, Formal analysis. Marianthi Markatou: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Software, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: AHT has received research support from Gilead Sciences, Novo Nordisk, AstraZeneca, and Salix paid to his institution. AHT has also received consultant and honoraria support from AbbVie, Gilead, Novo Nordisk, and Madrigal. AHT is also President of Empath Medical Innovations. MM is Vice-president of Empath Medical Innovations. DSF received research support from Gilead Sciences and Merck, paid to his institution. None of the other authors have additional declarations.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

Data availability

Data will be made available on request.